Confidential DeFi Platform
Three DeFi protocols, three different ways to lie to a user about their own money. My job was to make sure the UI never did.
2026
Timeline
UX/UI Designer & Frontend Developer
Role
The Problem
The platform needed to unify spot trading, lending, and perpetuals, each running on a different underlying protocol (Venus, Aster, Algebra, Carbon DeFi), into one Smart Wallet experience, while also making room for an AI agent to execute strategies on the user's behalf. The risk wasn't just visual inconsistency: mismatched states across protocols could put users in a position to submit a trade based on a price, balance, or permission that wasn't actually true.
The Challenge
The hardest problem wasn't any single screen: it was deciding what the UI should do when a protocol's data doesn't behave. Carbon's order book can be crossed with no honest mid-price to show; Withdraw needs to validate an amount and address live without silently blocking the user; a token or tab switch can leave a stale amount sitting in a field that now points at the wrong asset. Each of these is a small UI decision with real money behind it, so the fix was never 'add a null check'; it was deciding what's more honest to show: nothing, an explicit state, or a clear reason the action is blocked.
Goals
One wallet, three verticals
Give users a single, consistent interface for spot, lending, and perps instead of three separate protocol UIs.
Make agent execution legible
Design the simulate-then-confirm flow so handing a strategy to an AI agent feels transparent, not opaque.
Launch-ready trust signals
Treat price honesty, live validation, and state clearing as pre-mainnet blockers, not polish.
Background
This project, built by Protofire, is a non-custodial DeFi operating layer on BNB Chain: one Smart Wallet (a Safe) for spot trading, lending, and perpetuals, with AI agents able to execute strategies inside scoped, revocable permissions. It's built on real integrations with Venus, Aster, Algebra, and Carbon DeFi rather than a single in-house protocol, which means the frontend has to reconcile very different data shapes and failure modes under one consistent UI. The platform's name is withheld under an NDA.
My Role & Approach
I designed and built the core product screens (dashboard, positions, strategies, lending, and withdraw) and set up the navigation and theming system they share. As the project moved toward its first mainnet launch, I ran a dedicated UX/UI hardening pass: auditing every money-moving flow for validation gaps, stale state, and misleading data before real funds were on the line.
Process
01. System before screens
Set up navigation, theming, and the shared layout primitives before designing individual verticals, so Lending, Withdraw, and Strategies wouldn't each reinvent structure.
"I designed the navigation and theming system before the product screens, so I wouldn't redo layout for every new vertical."
02. Designing around protocol truth
Each integrated protocol surfaces state differently: Carbon's book can cross, Aster has funding rates, Algebra has LP ranges. The UI had to represent what was actually true, not paper over gaps with a plausible-looking default.
"I'd rather show nothing than show a fabricated price. If Carbon's book was crossed, the UI says so instead of making up a mid-price."
03. Validate live, don't block silently
Withdraw's amount and address checks moved from silent gating to live, explained validation, so users understand why an action is or isn't available.
"I chose to validate live and explain why, not just disable a button with no feedback."
04. The pre-mainnet sweep
Ahead of launch, audited every money-moving flow on its own terms (Withdraw, Lending, Perps), catching state leaks like an amount field surviving a token or tab switch.
"Before mainnet, I audited every money flow individually: a field that survives a token switch is a money bug, not a visual nitpick."
Outcome
Shipped the UI for the platform's core verticals and closed out a pre-mainnet bug sweep spanning Withdraw, Lending, and Carbon's spot pricing, including catching a case where the UI fabricated a mid-price when Carbon's order book was crossed, and fixing amount fields that silently survived a token or tab switch.
Tools
